Refine Inferencer CVE-2026-92784

Exploitation requires that the attacker control (or have compromised) the data provider / backend API that the refine application consumes, so that crafted JSON property names can be returned. A developer must then use @refinedev/inferencer (versions through 7.0.0) and render an Inferencer-generated page, at which point the unescaped field names execute as JavaScript in the developer's browser. Limiting factors: this is a development-time scaffolding tool rather than a production runtime component; it needs a malicious/untrusted data provider (unlikely with trusted first-party APIs); and it needs developer interaction (UI:P) plus specific crafted input (AC:H/AT:P), which together substantially narrow realistic exploitation.

This is a code-injection flaw (CWE-94) in a developer-time scaffolding tool, and the real-world risk is materially lower than the 7.7 CVSS 4.0 headline suggests. The vendor vector CVSS:4.0/AV:N/AC:H/AT:P/PR:N/UI:P/VC:H/VI:H/VA:H already encodes heavy limiters: AC:H (crafted field names required), AT:P (a specific attack precondition - the attacker must control the data provider), and UI:P (a developer must render an Inferencer page). PR:N is technically correct because no authentication to the victim's app is needed, but that is offset by the practical requirement of controlling or compromising the backend API the refine app consumes. There is no CISA KEV listing, no EPSS score, and no public exploit code in the provided references (only the GitHub issue #7556, the repo, two source-line references, and the VulnCheck advisory) - so exploitation status is: no public exploit identified at time of analysis. The impact rating VC:H/VI:H/VA:H reflects arbitrary JavaScript running in the developer's browser context, not server-side compromise; the 'RCE' tag overstates this since execution is client-side in a dev browser, not remote OS code execution. Net: a legitimate but niche, precondition-heavy issue best treated as medium priority rather than a drop-everything emergency.

A developer scaffolds CRUD views with @refinedev/inferencer against a data provider that is attacker-controlled or has been compromised (for example a third-party or self-hosted API returning attacker-influenced schema). The API responds with JSON whose property names contain crafted JavaScript; Inferencer interpolates those field names unescaped into generated JSX, so when the developer opens the Inferencer page the injected script executes in their browser. No public POC is present in the provided data, though the vendor references pinpoint the vulnerable source lines.